Summary: I built an optimization tool that helps Southwest Airlines mitigate the impacts of medium- to long-term disruptions such as the Boeing 737 MAX grounding, COVID-19 pandemic, and pilot shortages. The app allows planners to balance the customer impact of flight cuts with operational constraints and financial considerations.
Outcome: The tool has been used for crises, right-sizing the schedule to better match customer demand, and "freeing up" aircraft to use as charters or higher-value opportunities. The app helped Southwest save well over $100M in its first year of use. In his 2019Q3 earnings call, CEO Gary Kelly said, "We told you we would adjust for the MAX and produce satisfactory financial results, and I think we under-promised. These results are stellar." On the same call, President Tom Nealon added, "But even with the MAX challenge, we generated record third quarter passenger revenues... And I think our Network Planning Team, as Gary alluded to, has just done an incredible job of adjusting and republishing our schedules multiple times." The project was a semi-finalist for the 2020 INFORMS Franz Edelman Award and was featured at Southwest's 2019 Data Science Fair. Southwest named a plane for the Network Planning group after we won the company's Heroes of the Heart Award.
Summary: I delivered modeling that helps a large juice company optimize its supply chain - from the farm through to shipping finished goods from the processing center.
Outcome: With the optimization tooling, the company can better schedule raw goods arriving from farms to their storage silos on tanker trucks - resulting in less waste and demurrage cost. The tool accounts for strict food safety rules governing how silos must be emptied, cleaned, and filled with product. Given these constraints, the model also plans production in the processing center such that finished goods (customer orders) are completed on time.
Summary: I developed software that is used to schedule manufacturing processes and workers at a large electroplating facility.
Outcome: I worked with the client to formulate an optimization model and heuristic that mathematically solved the problem. The model was delivered to the client complete with a web-based interface and documentation, so schedulers can easily upload their inputs and get schedules out. The optimization app builds schedules that are 18% more efficient than those currently in use at the company - directly translating to substantially more throughput and less cost.
Summary: I developed two models for Muse Paintbar - one that optimizes customer seat assignments, and another that matches employees to the classes they're best-positioned to work.
Outcome: I worked with the folks at Muse to code their business problems into optimization models. I wrapped the models into Google spreadsheets so users could experiment with different inputs and give feedback on the modeling. "Andrew was a pleasure to work with - always making sure to communicate thoroughly, assist with any questions, and completed all the deliverables within the designated time frame," said Peter Levin, CTO at Muse Paintbar. "Andrew was a great partner on this project - thoughtful, thorough, and always pleasant. When we pursue our next optimization project, he will be the first person we seek out!"
Summary: I partnered with a colleague to build an app for portfolio managers at Dimensonal Fund Advisors that optimized how daily fund investments or divestments should be settled across a range of mutual fund products.
Outcome: The app was used internally to rebalance "fund of funds" products. The objective of the tool was to apply cash flows in a way that brings a given fund closer to the target weights of its sub-funds, given that it naturally becomes slightly out of balance each day when the market moves. The web app pulled fund positions and targets from a database each day and suggested how distributions could best be used to realign the fund across its various objectives.
Summary: I created a decision-making model that forecasts inventory needs and prioritizes daily production for an outdoor furniture manufacturer.
Outcome: The client sold an enormous range of products, which all had extremely variable and spiky order patterns. Despite the challenges, a strategy was established for forecasting, inventorying, and producing goods in response to orders and broad seasonal trends. The method was backtested using the orders of prior years. Desired business success metrics - such as customer service levels and overall inventory turn targets - were met or exceeded using the approach.
Summary: I built models at United Airlines that optimized staffing at airports and found creative ways to cross-utilize manpower within the operation.
Outcome: This set of tools helped United measure how different schedules would, at a high level, lead to cost implications for airports around the system. Additionally, the models provided suggestions for alternate shift patterns at airports that could better "cover" the work - both in the normal operation and when irregular operations occurred such as snowstorms or fog. These modeling efforts aided the company, as it was coming out of its merger with Continental Airlines, expand its operation without growing costs at the same rate.
Summary: I built a web app that created optimal rosters for fantasy basketball and generated the rosters of other contestants.
Outcome: Working with the client, a tool was built to generate rosters of basketball players based on the expected performance of those players and various constraints. The "most optimal" fantasy rosters tend not to provide the greatest financial returns due to repetition/overlap of top players online. To account for this, "the field" of other contestants' rosters were generated - against which the expected returns of these promising rosters could be measured.